Fault Detection Using Non-Linear Model-Based Approach

In this paper some practical issues related to fault detection implementation on a three tank process are presented. The importance of having some a-priori knowledge about the process, a mathematical model (derived from non-linear physical equations), and model- based FDI approach (using non-linear observers) is pointed out. FDI method tested in this paper was performed without the linearization of the model since the non-linear Luenberger observer enables a simple approach that does not require a lot of processing time. The implemented fault detection method proved itself well even when small abrupt changes/faults were generated on the real process. Two different methods for evaluation of the obtained residuals were applied using fixed and adaptive thresholds.